Auditable LLM-assisted decision support for sustainable infrastructure planning

Sustainable infrastructure decisions frequently rely on expert knowledge expressed in narrative form, creating a gap between qualitative judgement and the structured numerical inputs required by multi-criteria decision making methods. This study proposes an auditable large language model (LLM)-enabled decision-support workflow integrating the ordinal priority approach (OPA), a genetic algorithm (GA), and the technique for order preference by similarity to ideal solution (TOPSIS). Five construction-management experts provided memoranda on duration, cost, and environmental-impact trade-offs for a road project in Egypt. Five ChatGPT variants GPT-4.1, GPT-4.1-mini, GPT-4.1-nano, GPT-4o, and GPT-4o-mini were evaluated using identical prompts, temperature-0 decoding, no external retrieval, numerical validation, and expert review of rationale fidelity and weight consistency. OPA identified GPT-4o as the most aligned model, with a priority of 0.256. The selected LLM generated criterion weights for TOPSIS, while the GA produced 500 dataset-feasible alternatives. Using weights of 0.34, 0.33, and 0.33 for duration, cost, and environmental impact, respectively, TOPSIS used the LLM-derived criterion weights to distinguish among 500 feasible alternatives across the three decision criteria. The workflow provides a transparent and reviewable pathway from narrative expert judgement to quantitative infrastructure decision support, strengthening the auditability of LLM-enabled multi-criteria decision processes.